An overview on parametric quantile regression models and their computational implementation with applications to biomedical problems including COVID-19 data. (June 2022)

• Record Type: Journal Article
• Authors: Mazucheli, Josmar; Alves, Bruna; Menezes, André F.B.; Leiva, Víctor
• Is Part Of: Computer methods and programs in biomedicine. Volume 221(2022)
• Publication Date: 2022-06
• Subjects: GLM -- Likelihood methods -- Parametric regression -- Quantile function -- R software; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2022.106816
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: We provide a new computational package that includes 23 distributions for parametric quantile regression. We consider two biomedical applications from Brazil and US, one of them with COVID-19 data. We present a full up-to-date overview of parametric quantile regression. We implement a new R package. The package includes estimation, diagnostics, and several other statistical features for quantile regression. Abstract: Quantile regression allows us to estimate the relationship between covariates and any quantile of the response variable rather than the mean. Recently, several statistical distributions have been considered for quantile modeling. The objective of this study is to provide a new computational package, two biomedical applications, one of them with COVID-19 data, and an up-to-date overview of parametric quantile regression. A fully parametric quantile regression is formulated by first parameterizing the baseline distribution in terms of a quantile. Then, we introduce a regression-based functional form through a link function. The density, distribution, and quantile functions, as well as the main properties of each distribution, are presented. We consider 18 distributions related to normal and non-normal settings for quantile modeling of continuous responses on the unit interval, four distributions for continuous response, and one distribution for discrete response. We implement an R package that includes estimation and model checking, density, distribution, and quantile functions, as well as random number generators, for distributions using quantile regression in both location and shape parameters. In summary, a number of studies have recently appeared applying parametric quantile regression as an alternative to the distribution-free quantile regression proposed in the literature. We have reviewed a wide body of parametric quantile regression models, developed an R package which allows us, in a simple way, to fit a variety of distributions, and applied these models to two examples with biomedical real-world data from Brazil and COVID-19 data from US for illustrative purposes. Parametric and non-parametric quantile regressions are compared with these two data sets.